This repository constructs a rule-based framework for autonomous driving, including:
- path planning based on spline and Bezier curve
- velocity profile based on quintic polynomial optimization
- linear MPC and PID controller
- IDM following model and Mobile lane-changing model
- run ./CarlaUE4.sh
- run main_single_agent.py or multi_agent_agents.py
-
Environment Configuration:
- world_single_agent.py: Only ego car is controlled by the algorithm, and zombie cars are controlled by autopilot of carla.
- world_multi_agents.py: All the car are controlled by the algorithm.
-
Perception:
- feature.py extracts information from perception, including waypoints, lane points, traffic light, surrounding cars.
-
Trajectory Planning:
-
path_planner.py plan the path based on extracted waypoints
- bezier_curve.py used in lane changing task
- cubic_spline.py used in following task
-
velocity_planner.py formulates a quadratic program problem to optimize the quintic polynomial
-
path_planner.py plan the path based on extracted waypoints
-
Control:
-
pid.py: Throttle is calculated based on pid controller for velocity error, and steering is calculated based on pid controller for yaw and track error.
-
linear_mpc.py: The formulation of MPC is as follows:
where
$x_t=(p_{t,x},p_{t,y},v_t,\phi_t)$ ,$u_t=(a_t, \theta_t)$ ,$du_t=u_t-u_t$ . And$x_{ref}$ is obtained from path and velocity planner.$Q,R$ are the weights of MPC objective. The dynamic model is bicycle model that is defined in vehicle_model.py:
-
-
Rule-based decision:
- rule_decision.py: please note the defined status for vehicles to complish follwing, lane changing and stopping tasks. The details of IDM and Mobile model can be seen in websites listed in reference.
